| Gas Insulated Substation (GIS) has developed very quickly and has been in use all over the world, its compact size, high reliability, low maintenance have made it an attractive option in many circumstance. However, operation experiences show that intrinsic defects in GIS still cause accidents though its high reliability. It is well known that insulation breakdown is often preceded by Partial discharge (PD) activities. GIS internal defects can be discovered in time with the interferences such as corona discharge from the power system using Ultra-high frequency (UHF) method. Several types of defects occur in GIS, such as protrusion fixed to the conductor, free particles, surface contamination on the insulator and electrode gap defects and so on. Therefore, the analysis of UHF PD pulse shape, establishment of UHF PD mathematical models and pattern recognition of UHF PD are very important and necessary to realize UHF PD transmission, detect UHF PD signals, estimate GIS insulation condition and types of defects and guide its maintenance.In this paper, UHF PD signals of typical insulated defects in GIS are deeply studied based on analyzing researches about PD detection in GIS, establishment of mathematical model and its pattern recognition home and abroad; secondly, mathematical models are established, analyzed and justified; lastly, a new method of PD pattern recognition are introduced. The main achievements are as follows,â‘ A plenty of PD data are sampled with artificial physical models of PD defect in laboratory in different condition, including different defect size or shape, different detector location, different voltage level and so on, and the same time, UHF PD signals are analyzed roundly;â‘¡Mathematical models based on these defects are established, and the principles and methods of establishment are summarized. Furthermore, the parameters of mathematical models are given and the validity of PD mathematical model is proved by fitting error, power spectrum and time-frequency analysis. In the end, applications of the UHF PD mathematical models and complex wavelet transform for extracting the relevant signals from a white noise background are illustrated.â‘¢A new method of feature extraction for UHF PD signals based on the dual-tree complex wavelet transform is proposed with its shift invariance. Furthermore, feature space of UHF PD signals are constructed by energy values, maximum of module and statistical parameters, which describes the characters of UHF signals very well. At last, the validity of feature space has been proved by UHF PD mathematical models in the simulation.â‘£A new method of dimension reduction for feature space of UHF PD signals based on manifold learning is proposed, and LLE algorithm, ISOMAP algorithm, Laplacian Eigenmap algorithm and KPCA condense data from the high-dimensional space to the low-dimensional space and maintain the structure of the original sample patterns greatly. The results show that the UHF PD pattern recognition is satisfied when the sub-space features of energy, maximum values and statistical parameters are used as the input of the combined NN.
|
PDF Full Text Request